Person Case Constraints and Feature Complexity in Syntax

Person Case Constraints and Feature Complexity in Syntax Thomas Graf mail@thomasgraf.net http://thomasgraf.net Stony Brook University University of Vienna January 9, 2014

What is the PCC? Person Case Constraint (PCC) Whether the direct object (DO) and the indirect object (IO) of a clause can both be cliticized is contingent on the person specification of DO and IO. (1) Roger me/le leur a presésenté. Roger 1sg/3sg.acc 3pl.dat has shown Roger has shown me/him to them. Questions & Goals What are the descriptive properties of PCCs? algebraic unification in terms of presemilattices Can those properties be tied to independently motivated linguistic assumptions? connection to feature geometry

Outline 1 Person Case Constraints: An Overview PCC Typology Previous Proposals 2 Characterizing the Class of PCCs The Generalized PCC Algebraic Characterization via Person Locality 3 Connection to Feature Complexity Reducing Person Locality to Feature Complexity Reducing Feature Complexity to Feature Geometries 4 Another Look at the G-PCC

The PCC: A Closer Look attested in a variety of languages, including French, Spanish, Catalan, and Classical Arabic (Kayne 1975; Bonet 1991, 1994) specifics of PCC differ between languages, dialects, idiolects Four Attested PCC Variants Strong PCC (S-PCC; Bonet 1994) DO must be 3. Ultrastrong PCC (U-PCC; Nevins 2007) DO is less local than IO (where 3 < 2 < 1). Weak PCC (W-PCC; Bonet 1994) 3IO combines only with 3DO. Me-first PCC (M-PCC; Nevins 2007) If IO is 2 or 3, then DO is not 1. 1

The Four PCC Variants (Walkow 2012) IO /DO 1 2 3 1 NA * 2 * NA 3 * * NA (a) S-PCC IO /DO 1 2 3 1 NA 2 NA 3 * * NA (c) W-PCC IO /DO 1 2 3 1 NA 2 * NA 3 * * NA (b) U-PCC IO /DO 1 2 3 1 NA 2 * NA 3 * NA (d) M-PCC 2

The PCC in Minimalism Variety of proposals, work well empirically: Anagnostopoulou (2005) Nevins (2007) Béjar and Rezac (2009) Walkow (2012) Shared Idea: PCCs epiphenomenal, arise from more basic restrictions on the Agree operation Conceptual Drawbacks non-standard Agree mechanisms highly specific assumptions about feature system technical, complex hard to determine which assumptions are really needed 3

Example: Intuition Behind Nevins (2007) v[f ] vp VP IO V V DO v needs to agree with a particular feature f a search domain is established, depending on the type of f ungrammatical if the domain contains DO but not IO v agrees with both DO and IO IO and DO must have the same value for f 4

Example: Assumptions of Nevins (2007) Operations Agree steps happen concurrently constraints on search domain matching condition on IO and DO Structure clitics are PF-realization of Agree IO structurally higher than DO Features features are binary valued novel definition of contrastive features feature values can be marked or unmarked specific feature decomposition of person: Person Feature Matrix 1 [+author,+participant] 2 [-author,+participant] 3 [-author,-participant] 5

Evaluation Previous accounts work on an empirical level. They are complex because they try to do two things at once: 1 enforce the PCC with Minimalist machinery, 2 capture the attested typology. But that s more ambitious than necessary! The Secret Power of Merge (Graf 2011; Kobele 2011) Every syntactic constraint that can be computed with a finite amount of working memory can be enforced purely via Merge. The PCCs can be enforced by Merge, we do not need to extend our framework at all. The big issue is Point 2: There are 2 6 = 64 logically possible PCC variants. Why do we find only 4 PCCs? 6

The Generalized PCC The U-PCC was defined in terms of person locality. This system can be extended to all four PCC-types. Generalized PCC (G-PCC) IO is not less local than DO (IO DO), where S-PCC: 1 > 2 1 > 3 2 > 1 2 > 3 U-PCC: 1 > 2 1 > 3 2 > 3 W-PCC: 1 > 3 2 > 3 M-PCC: 1 > 2 1 > 3 7

Person Locality Hierarchies 1 1 2 2 1 2 1 3 3 3 2 3 (a) S-PCC (b) U-PCC (c) W-PCC (d) M-PCC 8

Example 1: S-PCC 1 2 3 1 > 2 1 > 3 2 > 1 2 > 3 IO /DO 1 2 3 1 NA * 2 * NA 3 * * NA 9

Example 2: W-PCC 1 2 3 1 > 3 2 > 3 IO /DO 1 2 3 1 NA 2 NA 3 * * NA 10

Presemilattices The G-PCC gives a unified description of the four PCCs, but we could have drawn any kind of graph. What makes the previous four structures so special? First, they are all presemilattices (Plummer and Pollard 2012). Definition (Presemilattices for Linguists) A structure S is a presemilattice iff for all nodes u and v of S, there is some node t such that t reflexively dominates u and v, or u and v reflexively dominate t. 11

Two More Restrictions The number of presemilattices with three nodes is still more than 4. We have to stipulate two more properties: Top and Bottom Top For all x, 1 < x implies x < 1. Every person feature is at most as local as 1. Bottom There is no x 3 such that x < 3. No person feature is less local than 3. Unifying the PCCs The class of attested PCCs is given by the G-PCC IO DO such that < defines a presemilattice P over {1, 2, 3}, and P respects both Top and Bottom. 12

Top and Bottom Match Feature Complexity Top and Bottom are stipulations, but express a common intuition: 1 is maximally complex, 3 minimally complex. Example 1: Person Specifications in Nevins (2007) Person Specification 1 [+author,+participant] 2 [-author,+participant] 3 [-author,-participant] Example 2: Alternative Specification a la Nevins (2007) Person Specification 1 {participant,author} 2 {participant} 3 {} 13

Doing Away with Top and Bottom? Syntactic proposals use feature geometry to derive PCC typology. Can we do the same? Yes, and No. Algebraic Feature Complexity [Idea Sketch] PCC locality is partially determined by feature complexity: Person features are ordered by their internal complexity algebraic structure C PCC locality rankings are exactly those structures that can be obtained from C by a map f such that f preserves certain properties of F The above is feasible, but more stipulative than one would expect. 14

What does C Look Like? C must assign different complexity to 1 and 2: * * 1 2 1 2 2 1 3 3 2 3 1 3 C must assign different complexity to 2 and 3: * * 1 1 1 2 1 3 2 3 3 2 3 2 15

The Only Viable Shape of C The previous arguments entail that C must be The 4 PCCs are generated from C by those maps that preserve maximality ( Top) preserve lack of daughter nodes ( Bottom) 1 2 3 But where does C come from? Can we obtain this complexity ranking from feature geometries? 16

Obtaining C from Feature Geometries C is easily obtained from the feature specification in Nevins (2007) if person complexity is determined by the number of features. Reminder: Set-Theoretic Specification a la Nevins (2007) Person Specification 1 {participant,author} 2 {participant} 3 {} This counting measure also works for the following specifications: Example: Specification with Distinguished Feature for 3 Person Specification 1 {participant,author,non-addressee} 2 {participant,addressee} 3 {non-participant} 17

Another Feature Geometry: Harley and Ritter (2002) Without restrictions on what counts as a complexity measure, any feature geometry can be the basis for C. But some feature geometries are compatible with more complexity measures than others. Example: Harley and Ritter (2002) Needs a Weighted Measure 1 and 2 are structurally equivalent in Harley and Ritter (2002): same number of features, same structural representation features must be weighted Person Specification 1 {ref,part,auth} 2 {ref,part,addr} 3 {ref} referring participant author addresse 18

Interim Summary The four PCC structures can be tied to feature geometries, but we need a complexity measure that obtains C from the geometry, and stipulations on how C restricts the class of PCC structures. In isolation there s many possible solutions, so at this point we cannot narrow things down further without looking at new data (gender, number, animacy). 19

Why IO DO? Reminder: Unifying the PCCs The class of attested PCCs is given by the G-PCC IO DO such that < defines a presemilattice P over {1, 2, 3}, and P respects both Top and Bottom. Maybe our problem with reducing the PCCs to feature geometries is due to our peculiar choice of G-PCC? Spoiler It is not. 20

Typology with Other Constraints a b c d IO DO S U W M DO < IO W U S M2 Me-second PCC (M2-PCC): If there is a DO, IO must be 1. [unattested] Under IO DO, M2-PCC is given by 1 2 3 Weakening Bottom to allow for this structure also brings in 1 2 3 21

Typology with Additional Structures a b c d e f IO DO S U W M M2 I DO < IO W U S M2 M N Indiscriminate PCC (I-PCC): No IO-DO clitic combinations. [Cairene Arabic (Shlonsky 1997:207, Walkow p.c.)] Null PCC (N-PCC): Any clitic combination. 22

The Full Extended Typology a b c d e f IO DO S U W M M2 I DO < IO W U S M2 M N IO < DO W U S M2 M N DO IO S U W M M2 I Implications Choice of G-PCC has minor effect on predicted PCC typology. Allowing structures e and f requires a change to Bottom/Preservation of lack of daughters. However, the complexity ranking C stays the same problem of linking C to feature geometry unchanged. 23

Technical Summary Fairly natural algebraic characterization of the attested PCCs: a ban against specific person locality configurations (G-PCC), locality structures must be presemilattices, locality structures respect both Top and Bottom. Going one level deeper: person complexity must be 1 > 2 > 3, person complexity restricts shape of locality structures (stipulative right now, but algebraically fairly natural). Going down another level: person complexity determined by feature geometry no obvious natural link at this point, but some geometries derive person complexity more easily 24

What s Next At this point there s too many algebraic solutions. We need to look at morphosyntax beyond person, i.e. number, gender, animacy. Ideally, all phenomena will follow naturally from a given feature geometry if all parameters have been fixed (mapping from feature geometry to complexity structures, mappings from complexity structures to locality structures). 25

References References I Anagnostopoulou, Elena. 2005. Strong and weak person restrictions: A feature checking analysis. In Clitics and affixation, ed. Lorie Heggie and Francisco Ordoñez, 199 235. Amsterdam: John Benjamins. Béjar, Susana, and Milan Rezac. 2009. Cyclic agree. Linguistic Inquiry 40:35 73. Bonet, Eulàlia. 1991. Morphology after syntax: Pronominal clitics in Romance. Doctoral Dissertation, MIT, Boston, MA. Bonet, Eulàlia. 1994. The Person-Case Constraint: A morphological approach. In The Morphology-Syntax Connection, number 22 in MIT Working Papers in Linguistics, 33 52. Graf, Thomas. 2011. Closure properties of minimalist derivation tree languages. In LACL 2011, ed. Sylvain Pogodalla and Jean-Philippe Prost, volume 6736 of Lecture Notes in Artificial Intelligence, 96 111. Heidelberg: Springer. Harley, Heidi, and Elizabeth Ritter. 2002. Person and number in pronouns: A feature-geometric analysis. Language 78:482 526. Kayne, Richard S. 1975. French syntax: The transformational cycle. Cambridge, Mass.: MIT Press. 26

References References II Kobele, Gregory M. 2011. Minimalist tree languages are closed under intersection with recognizable tree languages. In LACL 2011, ed. Sylvain Pogodalla and Jean-Philippe Prost, volume 6736 of Lecture Notes in Artificial Intelligence, 129 144. Nevins, Andrew. 2007. The representation of third person and its consequences for person-case effects. Natural Language and Linguistic Theory 25:273 313. Plummer, Andrew, and Carl Pollard. 2012. Agnostic possible world semantics. In LACL 2012, ed. Denis Béchet and Alexander Dikovsky, number 7351 in Lecture Notes in Computer Science, 201 212. Heidelberg: Springer. Shlonsky, Ur. 1997. Clause structure and word order in Hebrew and Arabic. Oxford: Oxford University Press. Walkow, Martin. 2012. Goals, big and small. Doctoral Dissertation, University of Massachusetts Amherst. 27